DataBench: leaderboard
Real-world tabular question-answering benchmark over many datasets, used in SemEval 2025 Task 8.
Metric: Score (%). Source: hex.tech. Status: years away from saturation. 32 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 (High) | 88 |
| 2 | Claude Fable 5 (Max) | 85 |
| 3 | Claude Opus 5 (Medium) | 83 |
| 4 | Claude Opus 5 (xHigh) | 80 |
| 5 | Claude Fable 5 (Medium) | 76 |
| 6 | GPT-5.6 Sol (xHigh) | 75 |
| 7 | Claude Opus 4.8 (Max) | 74 |
| 8 | GPT-5.6 Luna (xHigh) | 73 |
| 9 | GPT-5.6 Sol (High) | 72 |
| 10 | Claude Fable 5 (Low) | 72 |
| 11 | Claude Opus 4.8 (Low) | 72 |
| 12 | GPT-5.6 Sol (Medium) | 71 |
| 13 | Claude Fable 5 (High) | 71 |
| 14 | Claude Opus 5 (Max) | 70 |
| 15 | Claude Opus 4.8 (High) | 70 |
Interactive version: theaggregate.ai/benchmark?slug=databench · How It Works · Data refreshed daily, snapshot 2026-09-05.